Trustworthy evaluation of machine learning
Leakage-free dataset splits, cross-dataset testing, calibrated uncertainty and conformal prediction, so that reported accuracy reflects field performance.
Related serviceCompany
Liformatics is a technology consulting, research and development company based in Punjab, India. We work on problems where the answer is not yet known and has to be found by experiment before it can be built.
Mission
Useful methods in machine learning, security, robotics, quantum computing and optical wireless communication appear in research literature years before most organisations can use them. The obstacle is rarely the idea. It is the engineering between a published method and a system that runs every day on real data, inside real constraints.
Our mission is to do that engineering. We assess whether a method fits the problem, test it against a baseline, build it to production standard and hand it over in a form the client can operate without us.
We work from Vill. Taunsa in Nawanshahar district, Punjab, India, with CTOs, R&D heads, founders, government buyers, universities and industrial clients in manufacturing, healthcare, agriculture and telecom. The five practices are described on the services page.
Approach
The same method applies whether the work is a two-week review or a year-long build.
A feasibility question is answered with an experiment and a baseline. We commit to a full build only after the prototype meets criteria that were written down in advance.
Scope, assumptions, evaluation protocols, threat models and architecture choices are recorded in short documents that you review. Anyone joining the project later can see why a decision was made.
Work is released in milestones with a rollback path. Each stage ends at a point where you can stop, change direction or continue.
Source code, infrastructure definitions, tests and runbooks are delivered as part of the work, so that your team is able to maintain the system.
Values
Research focus
Client work and internal study feed each other. These are the topics our engineers return to between projects.
Leakage-free dataset splits, cross-dataset testing, calibrated uncertainty and conformal prediction, so that reported accuracy reflects field performance.
Related serviceAdversarial robustness, intrusion detection with learned models, and protection of IoT and industrial networks.
Related servicePerception and navigation for unstructured environments such as farms and mixed indoor sites, and aerial mapping of crop condition.
Related serviceBenchmarking of variational quantum algorithms against classical baselines on current hardware, and migration to post-quantum cryptography.
Related serviceLink design and resource allocation for networks that combine Li-Fi with radio, including indoor and greenhouse settings.
Related serviceMulti-objective and nature-inspired optimisation applied to model selection, controller tuning and network design.
Related serviceThe Liformatics engineering team works across all five service areas. Work is assigned by problem, not by department: a robotics engagement draws on the engineers who build perception models and on those who review its security, and a Li-Fi design is checked by the people who will test the network around it. Engineers review each other's designs and code, and the engineer who scopes a project stays with it through delivery.
Engineer
Engineer
Engineer
More about what we do and how to work with us.
The five practices, their deliverables and engagement models.
View servicesTypical applications in the six sectors we serve.
View industriesTechnical articles written by our engineers.
Read insightsThe skills we work with and how to send a CV.
View careersSend us the problem in a few lines. We will answer plainly, including when the answer is no.